An employee saved their company half a million dollars in a few minutes by making a simple change in the configuration panel. The situation was absurd, as the inefficiency they fixed should have been addressed much earlier. The company had a history of poor decision-making, hiring both talented individuals and inexperienced consultants, resulting in a chaotic and inefficient analytics platform. Despite the employee’s efforts, the organization seemed more interested in maintaining the illusion of competence rather than acknowledging its operational shortcomings.
This blog describes the experience of trying to add Python 3.12 support to a Python package and the challenges faced in the Python packaging ecosystem. It highlights the issues that come up when trying to use python setup.py and covers modern tools and standards like pyproject.toml. There are many issues with documentation and information for guidance on Python packaging.
Mixing client and server code is the new paradigm in React with Server Components. With the “use server” directive, you can run components exclusively on the server. This means that you can do things like write asynchronous database queries right in the component code. This post offers some thoughts on the benefits and drawbacks of this architecture.
Array.every() calls return true for empty arrays. As a result, it’s useful to consider whether there are any items in an array that don’t match a condition with every() call.
Odoo Experience, starting on October 8, promises to reveal Odoo 17, thought to disrupt the industry landscape once again. Anticipate innovative features, including a new website builder and a game-changing point of sale system adapted for restaurants.
Elon Musk wants to turn X into a financial hub that will fully replace the needs for banks within a year. Musk had plans to create an online bank called X with PayPal back in July 2000, but his plans never came to fruition. X aims to allow money to flow as freely as information or conversation on the platform. The company has already secured first money transmitter licenses in several states. It aims to make the service run globally.
Google has agreed to invest up to $2 billion in Anthropic. The deal involves a $500 million upfront investment with an additional $1.5 billion to be invested over time. Google already had a stake in Anthropic after investing $300 million in the company back in April. Anthropic was valued earlier this year at $4.1 billion.
Jeff Bezos recently showed off low-fidelity mock-ups of Blue Origin's Blue Moon lander to NASA officials. The Blue Moon lander is designed to deliver up to 3 metric tons of cargo anywhere on the lunar surface. Blue Origin will be responsible for transporting astronauts between lunar orbit and the surface of the Moon and back to space on the Artemis V mission scheduled for 2029. Pictures of the lander are available in the article.
Eukaryotic cells have membrane-bound nuclei and a system of internal membranes called the endoplasmic reticulum. There are several competing theories about how these structures developed. Scientists have only recently been able to start investigating the origin of these structures using new technologies. This article provides an overview of the theories around the structures and a roadmap of how scientists plan to uncover their beginnings.
This article lists 11 things Site Reliability Engineers at Google have learned from 20 years of working in the field. Google is now over 1,000 times as large as it was 20 years ago, with a network 10,000 times larger. The company now spends far less effort per server than it used to while enjoying much better reliability. Its tools have evolved throughout the years into a unified platform that offers reliability by default.
ScratchDB is an alternative to BigQuery, Redshift, and Snowflake that runs on Clickhouse. It lets users input arbitrary JSON and perform analytical queries against it. ScratchDB automatically creates tables and columns when new data is added.
Google's senior vice president Prabhakar Raghavan testified during the company's monopoly trial, saying that Google's default agreements with mobile phone manufacturers and web browsers were the company's biggest cost in 2021. Google paid $26.3 billion for default agreements while raking in $146.4 billion in revenue from search advertising in 2021. The company says that these deals are legal and that its dominance is not due to them.
People are now having hours-long discussions with ChatGPT on the go. The latest ChatGPT update allows the AI to see, hear, and speak. It enables people to use the assistant in many more situations for things like creative development, brainstorming, or simple conversation. Being able to interact with the assistant through voice has made more people feel a deeper emotional connection with it, with many people drawing similarities between ChatGPT and Samantha from the 2013 movie 'Her'.
This page contains a collection of tactics, techniques, and procedures of offensive ML attacks that focuses heavily on attacks that have code that can be used to perform them right away.
Reproducible Builds provide a reliable way to verify that binaries are faithful to their sources and have not been tampered with anywhere in the build pipeline.
US District Judge Lewis Kaplan has ruled that Sam Bankman-Fried cannot provide testimony that he relied on lawyers to steer his decision-making in his case to the jury.
Amazon Ads has introduced an AI-powered image generation feature in beta that creates lifestyle and brand-themed images to improve ad performances. The tool, which requires no technical expertise, enables brands to create more engaging ads, reportedly increasing click-through rates by 40%. It will be rolled out progressively to advertisers based on customer feedback.
Emulator Fine Tuning (EFT) comes from the direct preference optimization group and explores what happens if you fine tune a small model and then project that fine tuning onto a large model. This counterintuitive strategy works shockingly well.
ML models can sometimes wrongly recognize unfamiliar data, thinking they've seen it before. A new method, NNGuide, helps models make better decisions, especially when identifying unknown data.
Researchers use sparse autoencoders to tackle neural networks' polysemanticity, extracting clearer features for improved model understanding. These features influence outputs and exhibit complex interactions.
This study presents "JudgeLM", a new method to efficiently evaluate Large Language Models in versatile situations. The authors developed a vast dataset and benchmarking system.
The new GPT-4V allows for interactive conversations with image inputs but also introduces new vulnerabilities through visual prompt injection attacks, enabling exfiltration of data and execution of hidden instructions encoded in images.
Rust is a popular systems programming language with strong memory safety. Burn is a framework built on these principles for efficient ML. Importantly, it offers compatibility with the Onyx runtime.
OpenAI's prompts reveal many quirks about prompt engineering. For example, the prompts use 'please' a lot and important instructions are often all capitalized. Capitalizing instructions likely works as the models were trained on huge numbers of examples where responses clearly paid more attention to capitalized sentences. This could result in a future where people shout at their computers to get them to work better.
AI alignment is a complex issue that lacks concrete benchmarks or numbers for measurement. Predictions of AI-induced doom rely on narratives rather than rigorous models. The absence of hard numbers makes it challenging to assess the state of AI alignment.
The Llava fine tuning paradigm combines powerful language models with vision capabilities. It has been applied with Mistral 7B for improved performance.
The Dynamic Scooter Model B [5] might be the smallest and lightest e-scooter out there. Join the oversubscribed Kickstarter [5] for early bird pricing and many other goodies.
SEC Commissioner Hester Peirce has expressed disagreement with the SEC's recent case against crypto company LBRY, arguing that its enforcement-driven approach to the crypto industry is flawed. She questions whether the closure of LBRY, an operational company with real-world blockchain applications, actually benefits investors or the market. For context, Peirce has consistently advocated for a spot bitcoin ETF and leniency towards crypto, emphasizing the need for clear regulations before enforcement.
Gemini Trust filed a lawsuit against bankrupt Genesis Global Holdco and DCG, claiming it should have access to GBTC shares worth around $1.6 billion. It alleges that these shares were promised as collateral for loans by Gemini users in the Gemini Earn Program. Furthermore, Gemini argues Genesis has hurt users by delaying the return of their assets. Both firms are facing regulatory scrutiny, while Genesis' lending unit has filed for bankruptcy.
Arbitrum has officially launched Arbitrum Orbit on mainnet, an advanced scaling technology for launching customized blockchains. Orbit chains include Xai Games, Cometh, SyndrHQ, Polychainmon, and others forming an expanding ecosystem. Orbit provides solutions for ultra-low-cost transaction fees and smart contracts in Rust, C, and C++ with full EVM compatibility. The technology will also allow for future upgrades, including validation, MEV capture, cost reductions, and more.
Neo announced a new EVM-compatible sidechain aimed at empowering developers to deploy smart contracts and expand DeFi use cases. The sidechain focuses on countering MEV attacks to maintain platform integrity and prevent user losses. Neo aims to offer a secure and reliable DeFi experience with transparency of transaction metadata and identity protection. The sidechain remains connected to the Neo Mainnet through Neo's native GAS token.
This thread goes over the five stages of the Bitcoin Halving set for April 2024, including how Bitcoin typically trades around its halvings. These stages include the pre-halving period, the pre-halving rally, pre-halving retrace, re-accumulation, and the parabolic uptrend. Historically, retraces during the pre-halving have yielded great returns post-halving. The re-accumulation stage often tests investor patience, yet leads to the uptrend stage that typically sees Bitcoin reach record highs.
Numerous rollup frameworks and Rollups-as-a-Service platforms have been launched, leading many to wonder about their viability for value accrual. Although direct value primarily accumulates at the execution layer, rollup frameworks can acquire indirect value through widespread use and enhancements. Some may also share revenue, facilitating semi-direct value accumulation. Striking the right balance between collaboration and competition will be key for these frameworks and platforms' sustainable growth.
ETH Hong Kong highlighted how Hong Kong is emerging as a hub for crypto investment and startups amidst China's economic and regulatory shifts. A lot has changed since 2020. Many of the key players have shifted. The macro environment has significantly changed, with some places like Hong Kong embracing the technology while other countries lag behind in regulations.
Bitcoin surged to $35,000 last week, leading to the highest spot volume trading day since spring at above $24 billion. This price increase was attributed to more bullish news on a potential Bitcoin exchange-traded fund, which could attract additional capital into the space. Bitcoin’s price surge also spilled over into ETH, SOL, and other alts.
In this thread, Raoul Pal discusses how global liquidity affects crypto prices, noting many of the various liquidity measures he tracks are inflecting in positive directions.
NFT artists Ryder Ripps and Jeremy Cahen are being ordered to pay Yuga Labs $1.57 million plus legal fees for copyright and trademark infringement in a "copycat" NFT lawsuit.